Early Warning Pedestrian Crossing Intention From Its Head Gesture using Head Pose Estimation

Abstract

The development of the autonomous driving system is still a hot topic. Especially in the area of the interaction between the autonomous driving system and road crossing pedestrians. Recognizing the pedestrian crossing intention is one of the crucial topics for the smart autonomous driving system. The autonomous driving system must be safe enough for both its user and pedestrian. Many research, approach, and method has been developed to detect or predict pedestrian crossing intention. But unfortunately, many previous works predict the pedestrian crossing intention that already does cross the road. There is still few research about how to predict early pedestrian crossing intention. Usually when they want to cross the road or before they do cross the road, the pedestrian gives unique gestures like looking at the incoming vehicle. Thus, an early warning pedestrian crossing intention system has been developed using a different approach, by its head gesture. The pedestrian crossing intention could be predicted by classifying its head pose angle. Experiment show very well that the proposed method able to predict early pedestrian crossing intention by its head gesture and give an early warning sign to the autonomous driving system with the accuracy of the classification 97.2%. We believe our work could bring benefits to the autonomous driving system to increase the safeties of its system.